Is lack-of-fit of the model relevant in case there are few missing counts?
The imputed indices will be correct, but model based indices and slopes can be biased if the model fits badly. It is worth trying to find a better fitting model: the reward could be better model indices with smaller standard errors.
